[4] Another important feature of King's EI procedure is that it allows researchers to model aggregation bias (that violates one assumption of King's basic model) through the use of covariate(s) in extended models. To discover possible aggregation bias, EI provides useful graphic diagnostics (King, 1997: 158-196; 282-291). All estimates of precinct-level white crossover in the 29 elections under investigation were checked for possible aggregation bias based on King's recommended diagnostics. Extended models were employed for 18 of our elections. 

The EI software is provided by King through his web page at http://Gking.harvard.edu. The EI procedure first computes turnout rates for each racial group both at the election unit and at the precinct level. Three precinct-level variables are used to estimate white turnout – the number of registered votes, the proportion of black registered voters, and the proportion of registered voters who signed in to vote in each precinct. After white turnout is computed, the percentage of votes received by black candidates is used to calculate the level of racial crossover voting.
Back